Stephan

Just open-sourced PLAY. PLAY. is the most premium video player on Windows. It's 100% free and you can install it now.

@philippejeanpr1 @BayesReality On serait tenté de dire que les entreprises sont parfaitement rationnelles et que chaque poste est créateur de valeur. Et puis un type comme Musk débarque et vire la moitié de la masse salariale sans dégrader le produit, prouvant du même coup que les "Bullshit Jobs" existent.

@BayesReality Pour une fois pas d'accord. La proportion de parasites est élevée et l'administration publique ou privée sait très bien s'étendre sans nécessité réelle. Loi de Parkinson - malheureusement.

On est d'accord que les bullshit jobs c'est surtout un terme pour ceux qui sont pas assez intelligents pour comprendre comment marche la division du travail, ou la complexité d'une grande entreprise (ou la masse de bullshit est un side effect du coût de coordination)

Cool! I contributed to this skill, so seeing actual users get value from it feels pretty nice

We're too poor to run glorified ralph loops on our 20€ plan anyway 😄

I explored an alternate timeline where France defeated Britain at Trafalgar, and North America east of the Mississippi became permanent French territory. Over time, the continent evolved into a French-speaking world. Enjoy:

Seedance restrictions are getting bypassed and we're getting new AI episodes of The Office Credit: @kaikaiohwhy

@ccssmnn yes! this is the way!

ESLint >>> AGENTS.md

I showed my boss a POC agent for our internal app and he asked me if I could try with a local LLM. Tried it with Chrome’s Gemini Nano and wow! Same agent, no API call. I did have to build my own tool factory to make tool calls work, but wow. Local inference is getting real.

@MassAppeal "I break bread, ribs, 100 dollar bills"

what are the best opening bars in Hip Hop history? 🤔

@ncapeyronSABA @lafibreinfo @pbeyssac VLC et MPV décodent le AV1 sur à peu près tous les devices, Apple compris

@lafibreinfo @pbeyssac Huh ?? Bizarre cette affirmation sur Apple… Il me semble pas qu’Apple ait ban ce codec et n’importe quel player vidéo peut l’implémenter. VLC Mac le fait sans problème ainsi que d’autres. Et j’ai testé avec les vidéos de test sur mon iPhone avec Safari et ça passe :

Le codec #AV1 permet : - 45% d’interruptions en moins - 33% de bande passante économisée - Une qualité visuelle supérieure Sa généralisation se heurte à une incompatibilité d'une partie de l'écosystème #Apple, faut de décodeur logiciel AV1. ➡️ lafibre.info/tv-numerique-h…

@Cedric_r @512banque Très bon ! Pas sûr que la partie scoring serve à quelque chose (les LLMs adorent générer un scoring mais d'expérience ça n'apporte rien) mais sinon j'aime beaucoup le reste du skill ! Merci pour le partage

@512banque Et utiliser des skills pour enlever les tics de l’IA cedricrittie.com/fr/ressources/…

les gars écrivez vos posts avec de l'IA ya aucun souci, c'est le jeu, mais au minimum filez-lui du contexte pour lui apprendre comment vous vous exprimez naturellement. c'est un massacre sinon.

Fascinating article! 7 LLMs commanding ancient armies in a custom battle simulator: cavalry, elephants, morale breaks, the whole thing. GPT-5.5 wins by a large margin while being incredibly token-efficient. This matches what I’ve been seeing in day-to-day agentic tasks too.

@sarahmsachs The real good stuff is Deepseek V4 flash! It's so cheap you can do all sort of advanced token burning agentic patterns such as ralph loops, evals, chrome mcp, etc..

DeepSeek v4 works fine, but it’s not the frontier-pressing moment we saw with Kimi 2.6. On Notion eval data, it’s similar performance to GPT 5.2, with understandable failings. Most interesting — it doesn’t scale well. It’s ridiculously slow. On multiple major, trusted, and performant US inference providers we see it 15x slower than GPT 5.2 and 2x slower than Opus 4.7, a problem Kimi never had. Curious if it’s a fundamental issue in architecture, or a matter of time til inference providers make it work. Doesn’t seem urgent either way, if Kimi can outperform. Cheaper maybe, but not groundbreaking.

What’s really cool about models like DeepSeek V4 Pro and plans like @opencode Go is that they make advanced agentic patterns cheap enough to actually play with. Ralph loops, self-improving agents, eval loops, etc. Huge unlock.

This isn't really a surprise considering OpenAI wrote it its recent harness engineering paper that they implemented a Ralph Loop for their internal project openai.com/index/harness-…

The /goal command in Codex (and pi) is actually some sort of elegant implementation of the Ralph loop!

@micLivs @badlogicgames is it some sort of ralph loop?

Get ready to hit your goals (and usage limits) Introducing pi-goal, pi-goal is a pi implementation for the recent /goal command in codex, does the exact same thing. pi-goal adds a /goal command and goal tools so @badlogicgames's pi can keep working toward a long-running objective until the goal is complete, paused, cleared, or token-budget-limited.

Also I was building APKs through EAS free tier which was painfully slow. Now Codex compiles them locally in seconds. Installed dependencies, skipped Android Studio entirely and it just works. Should've done this weeks ago!

Spent way too long suffering through Expo Go so I asked Codex to hook itself to my Android via ADB. Now it iterates native builds directly on my phone, completely unsupervised. This is the way.

LinkedIn is basically Moltbook now.

Let's give this bad boy a spin then